In this paper, a novel robust watermarking research for color image based on the Genetic Algorithm(GA) and BP neural network is proposed which combine the advantages of BP neural networks and GA algorithms. During the detecting process, the GA aims to optimize the parameters of the BP neural networks and the optimum values are regarded as the initial values and weight matrix of the BP neural network parameters. Furthermore, we utilize the gradient descent learning algorithm to adjust the parameters to data fusion the watermark confidence. The experiments show that the watermarking algorithm is imperceptive and more robust than the BP neural networks method.